What it does This repo is a curated markdown index of datasets for computer-vision segmentation tasks: salient object detection, camouflaged object detection, video object segmentation, image matting, and a handful of industrial and medical niches. Each entry links to the original paper, project page, and download URL where still alive. The maintainer also tracks related resources in GitHub issues.

The interesting bit The breadth is the point. It spans RGB saliency classics (MSRA10K, DUTS), RGB-D and thermal variants, co-saliency, defocus blur detection, anomaly detection (MVTec AD), even trichomonas vaginalis segmentation. For a field where benchmark chasing lives or dies on train/test splits, having a single living index beats scattered paper appendices.

Key highlights

• 50+ dataset entries across 10+ sub-disciplines, from saliency to matting to medical imaging
• Each entry includes paper citation, original project URL, and direct download links where available
• Covers niche modalities: RGB-T (thermal), RGB-D, video, high-resolution, and open-vocabulary camouflage
• Active issue tracking for additional datasets the maintainer hasn’t yet merged
• Includes related survey papers and resource websites for tooling

Caveats

• Some download links point to Baidu Pan or personal lab servers; longevity is uncertain
• README mixes Chinese and English descriptions without consistent structure
• No code, no unified loader, no standardized format — this is purely a reference list

Verdict Worth bookmarking if you’re building or benchmarking segmentation models and tired of reconstructing the dataset landscape from paper citations. Skip it if you need a Python package with torch.utils.data.Dataset wrappers; this is a wiki, not a framework.